Method and device for inspecting hot glass containers with a view to identifying defects

Abstract

The invention relates to a method for inspecting glass containers (2) while they are still hot, the method consisting in, for each container: - acquiring at least one transmission image (It) of the container illuminated by a source (14) of light passing through the container, and at least one infrared image (Ir) of the container, - analysing at least one transmission image and at least one infrared image, - registering at least one portion of the transmission image and at least one portion of the infrared image, - classifying the container, based on at least one transmission image and at least one infrared image, which have been registered, in order to identify, for a container, at least one type of defect.

Description

Technical Field

[0001] The present invention relates to the technical field of online inspection of transparent or translucent containers such as, for example, glass bottles, jars or flasks, for the purpose of quality control in order to detect and identify any defects that may affect these containers.

[0002] The object of the invention finds particularly advantageous applications for analyzing the physical characteristics of containers in order to identify non-conforming physical characteristics corresponding to defects, such as surface defects, such as folds or crevices, internal defects in the material, such as cracks, inclusions, or bubbles, or dimensional defects of the container, such as the distribution of glass. Previous technique

[0003] For the manufacture of glass containers, the manufacturing process, which includes melting the glass and then conveying it to forming units, is carried out using a manufacturing plant comprising a melting furnace, a feed core for the molten glass, and a forming machine, generally of the type designated as an IS machine. The IS forming machine includes a distributor that forms droplets of glass called parisons and distributes them through conduits called "deliveries" to forming sections. At the exit of these sections, the containers are at a high temperature, typically between 300°C and 600°C. The containers, freshly formed by the forming machine, are placed successively on an outfeed conveyor to form a line of containers. These containers are then transported in a continuous line by a conveyor to various processing stations.In particular, the formed containers are placed in an annealing oven, which raises their temperature before cooling them in a controlled manner to eliminate the thermal stresses created by the forming process. Other glass container forming processes are known for tableware, insulators, syringes, and ampoules. For example, there are forming machines such as rotary and sequential presses, rather than parallel-section forming machines like the IS machines. There are also machines that transform preforms into tubes, particularly monocalcium glass, to produce syringes and ampoules for pharmaceutical products.

[0004] It is known to systematically inspect all containers exiting the annealing furnace using various inspection equipment, including wall inspection systems such as those described in patents EP2082216, EP2145175, or EP2856122. A known wall inspection technique is the transmission method, in which a light source is positioned on one side of the conveyor and at least one camera (typically 2 to 6, 12, or 24) is positioned on the opposite side to acquire at least one image formed by the light transmitted through the container walls. According to patent EP 1,109,008, a method for analyzing container images for cold inspection is described. A segmentation step detects features within the images and regions surrounding these features.Discriminatory parameters for a region are calculated and combined with a fuzzy logic method to determine the most probable type of feature from a list of features corresponding to possible defects. The region's conformity is then determined by applying different criteria depending on the type of feature selected. For example, a fold-type defect will be rejected for a certain surface area, while an inclusion-type defect will be rejected even if it has a small surface area. This patent notably teaches that not all defects have the same criticality, which justifies attempting to determine their nature before deciding to reject a container.

[0005] Generally speaking, it is beneficial to identify defects as early as possible in the glass container manufacturing process so that they can be corrected as soon as possible at the manufacturing plant level. It is therefore advantageous to detect defects in containers that are directly related to settings in the forming process, in order to correct any deviations in the forming process as quickly as possible, although this remains complex due to the non-linearity of the system, the number of process parameters, and the fact that a single cause produces several combined effects and that several causes can explain the same effect.

[0006] This is why prior art has proposed methods for observing still-hot containers moving along the outfeed conveyor of the forming machine, using devices similar to those employed for cold inspection after the annealing arc. For example, according to patents EP 0 177 004 and EP 3 516 377, wall inspection is carried out using the same methods as cold inspection, and these patents describe, in particular, ways of adapting the inspection to the specific environment of the hot sector by arranging means for managing the high ambient temperature. Thus, a light source located on one side of the path of the containers on the outfeed conveyor of the forming machine illuminates the containers, and at least one camera sensitive to the emitted light acquires at least one transmitted image of each container. The images are analyzed to identify containers with defects.

[0007] US patent 6,584,805 also describes an installation for observing still-hot containers moving along the outfeed conveyor of a forming machine. This installation includes an inspection station with light sources arranged along one side of the conveyor to illuminate the containers. This inspection station also includes cameras arranged along the other side of the conveyor to acquire transmitted images of the containers illuminated by the light sources. This inspection station allows, in particular, the measurement of the container diameters at different heights. These measured diameters are compared to reference diameters to determine whether the containers are defective.

[0008] Since hot containers emit infrared radiation, prior art also proposed methods for observing hot containers moving along the outfeed conveyor using an infrared camera, based on the principle that thicker regions of the containers radiate more. Patent EP 0 643 297 describes a device for performing analysis and diagnosis on a glass manufacturing process, comprising a sensor sensitive to the infrared radiation emitted by objects exiting the forming machine. This system also includes a digital processing device that compares the radiation to a mathematical reference model to determine deviations in the glass distribution and / or the causes of thermal stresses within the container.

[0009] Patent EP0 679 883 describes an image acquisition device using an infrared camera that collects the radiation emitted by hot containers as they exit the manufacturing machine. The infrared camera is synchronized with the operation of the container forming cavities. For inspection, processing software divides the container image captured by the infrared camera into inspection regions. By segmenting and measuring light or dark spots located within these regions, it is possible to identify trapezoidal defects and inclusions.

[0010] According to patent DE10030649, knowing the speed of the output conveyor and the order of exit of the containers from the different sections, it is possible and advantageous to link the detected defects to the original cavity in order to be able to act judiciously on the manufacturing process to correct the defects.

[0011] Prior art therefore teaches us to inspect the still-hot containers as early as possible in order to quickly adjust or correct the forming process. Dimensional checks of diameters or heights are performed by vision, and even, according to patent application WO2021009456, glass thickness can be measured by infrared imaging in at least two specific wavelength ranges. A dimensional deviation is an identified defect in itself—whether in height, diameter, inclination, or thickness—and can be attributed to at least one cause of drift in the manufacturing process.

[0012] US patent 6,188,079 proposes a method for measuring the glass thickness of a container using infrared radiation. This method involves measuring the intensity of the radiation in a spectral band where the radiation is emitted by the material between the container's external and internal surfaces. The first spectral band, whose signal depends on both the glass temperature and thickness, is preferably between 0.4 and 1.1 microns. The method also involves measuring the intensity of the radiation in a second spectral band where the radiation is emitted substantially entirely by a single external surface of the container. According to this patent, the second spectral band, where the radiation depends only on the temperature and corresponds to surface radiation, is preferably between 4.8 and 5 microns.The method consists of determining the thickness of the container between its external and internal surfaces as a combined function of the first and second measured intensities. In other words, the thickness and temperature are determined from the two radiation measurements taken in the first and second spectral bands.

[0013] The situation is different regarding what we will call "appearance defects," namely visual defects of any type: inclusions (of foreign bodies such as ceramics or metal), bubbles, folds, rivers (surface grooves), glazes (cracks), fins, trapezoids, grease spots, very thin areas, and unmelted areas. These appearance defects manifest in images as local optical variations or pixels that deviate from the background. These appearance defects can be critical if they lead to a risk to the consumer, a risk of breakage, or a loss of functionality of the container. However, recognizing an appearance defect from an image can be ambiguous. Consequently, due to the safety margins implemented during detection, some containers are considered defective even though they are compliant.

[0014] Furthermore, it should be noted that acceptable artifacts, such as engravings or decorations, and faint mold seams, can be distinguished in the images. Therefore, it is necessary to precisely identify the nature of surface defects in order to identify critical defects and differentiate them from other defects. Preventing critical defects requires improving the reliability of surface defect classification. Besides improving defect identification, this identification allows for the determination of their potential causes, enabling the manufacturing process to be controlled according to the category of detected defects. Indeed, without reliable identification of detected defects, no sound decision regarding corrective action in the manufacturing process can be made, whether manually or automatically. Description of the invention

[0015] The present invention aims to remedy the disadvantages of the prior art by proposing a method of quality control of hot glass containers, designed to achieve more effective detection of defects, particularly of appearance, while ensuring their safe and certain identification in order to give more complete information on the corrections to be made to the control parameters of a glass container manufacturing process in a manufacturing facility.

[0016] The object of the invention relates to a method for inspecting still-hot glass containers coming out of a manufacturing installation, in order to identify, for a container, a type of defect, the method consisting of the following for each container: to acquire at least one transmission image of the container illuminated by a light source passing through the container and at least one infrared radiation image of the container, to analyze at least one transmission image and at least one infrared radiation image, to ensure a matching of at least a part of the transmission image and at least a part of the infrared radiation image, to classify the container, from at least one matching transmission image and at least one matching infrared radiation image, in order to identify for a container, at least one type of defect.

[0017] According to an advantageous embodiment, the container is illuminated by a light source whose emission spectrum is in a wavelength range below 0.8 µm, and the infrared radiation image of a container is acquired in a wavelength range above 0.8 µm.

[0018] For example, an infrared radiation image of a container is acquired when the light source is turned off.

[0019] According to another implementation example, the infrared radiation image is acquired from an observation direction such that the light emitted by the light source is not captured with the infrared radiation from the container.

[0020] According to one feature of the invention, to ensure the matching of transmission images and infrared radiation images, the method detects candidate regions in the transmission images and in the infrared radiation images, the method ensuring, for each container: a matching of candidate regions of transmission images or candidate regions of infrared radiation images with the corresponding regions respectively of infrared radiation images and transmission images, according to their position on the container, or a matching of candidate regions of transmission images with candidate regions of infrared radiation images.

[0021] According to one embodiment, the process ensures, as a matching mechanism, a fusion of transmitted images and infrared radiation images to obtain a composite image, the process ensuring: an extraction of classification features from the composite image, expressing classification criteria in transmission and radiation, and a classification of the container using the classification criteria in transmission and radiation.

[0022] According to another embodiment, the process ensures, as a matching mechanism, a fusion of transmission images and infrared radiation images to obtain a composite image, the process ensuring: a segmentation of composite images to detect composite candidate regions, an extraction of classification features of composite candidate regions, expressing classification criteria in transmission and radiation, a classification of the container using the transmission and radiation classification criteria of the composite candidate regions.

[0023] According to one feature of the invention: We extract transmission images, transmission classification criteria; we extract infrared radiation images, radiation classification criteria; we classify the container using transmission and radiation classification criteria.

[0024] Advantageously, radiation classification criteria are chosen for infrared radiation images and transmission classification criteria for transmission images, and / or composite criteria that take into account characteristics combining transmission and infrared radiation images in a logical or mathematical way, these radiation and transmission classification criteria being criteria of position, size, shape or photometry.

[0025] Advantageously, the process consists of classifying the container using a supervised learning classifier whose input data are: the criteria for classification in radiation and transmission, or radiation images and transmission images, or parts of radiation images and parts of transmission images.

[0026] According to one example of implementation, the process consists of classifying the container by a supervised learning classifier whose input data is at least one composite image obtained by merging at least one radiation image with at least one transmission image of a container or by merging regions of at least one radiation image with corresponding regions of at least one transmission image.

[0027] According to another implementation example, the process involves classifying the container using a supervised learning classifier trained on a training database consisting of a set of records, each containing, for an observed example container: at least one radiation image of the container, at least one transmission image of the container and at least one label assigning to the example container at least one object class from a list of possible classes such as defect types, or at least one radiation image region of the example container, at least one transmission image region of the example container and at least one label assigning to the corresponding region of the example container at least one object class from a list of possible classes such as defect types.

[0028] According to an advantageous feature of the invention, the method aims to classify each container according to at least one class of object from a list of possible classes containing at least types of defects, the list of possible classes including at least: no defect, trapezoid, inclusion, bubble.

[0029] According to an advantageous feature of the process, a step is implemented to take into account at least one type of detected defect in order to deduce adjustment information for at least one control parameter of the manufacturing installation.

[0030] Another object of the invention is to provide a device for inspecting still-hot glass containers coming out of a manufacturing facility in order to identify, for a given container, a type of defect, the device comprising: a transmission image acquisition system for containers and infrared radiation image acquisition of containers, an information processing unit connected to the image acquisition system, this information processing unit being configured to include: * a system for analyzing at least one transmission image and at least one infrared radiation image of the container, * a system for matching at least one region of a transmission image and at least one region of at least one infrared radiation image of the container, * a classifier of the container, from at least one region of at least one transmission image and at least one region of at least one infrared radiation image, matched, in order to identify for a container, at least one type of defect.

[0031] According to an example embodiment, the system for acquiring transmitted images of containers and infrared radiation images of containers comprises, on the one hand, a camera sensitive to the infrared radiation emitted by the containers and equipped with a lens, and on the other hand, a source of light passing through the containers and a camera sensitive to the light transmitted by the containers and equipped with a lens.

[0032] Advantageously, the image acquisition system includes a system for selecting the light emitted by the light source and positioned to eliminate, from the radiation captured by the camera sensitive to infrared radiation, the light emitted by the light source.

[0033] For example, the system for acquiring transmission images of containers and infrared radiation images of containers includes: a light source illuminating the containers, a sensor sensitive to the infrared radiation emitted by the containers, a sensor sensitive to the light emitted by the light source and transmitted by the containers, a common optical objective for recovering the infrared radiation emitted by the containers and the light transmitted by the containers, this optical objective being associated with an optical separation and filtration system to eliminate the light emitted by the light source from the radiation received by the infrared-sensitive sensor

[0034] According to an implementation characteristic, the information processing unit is connected: to an ejector to control the ejection of containers identified as defective, and / or a display unit to present to an operator the identified defects, transmitted images and infrared radiation images of the containers.

[0035] Typically, the information processing unit is connected to a production computer that monitors the manufacturing installation in order to: receive from the production computer, time information enabling the association of containers, their images and their detected defects with the mold number or the forming cavity, transmit to the production computer, the identified defects and measurements taken, so that the production computer can automatically deduce adjustment information for at least one control parameter of the manufacturing installation. Brief description of the drawings

[0036] [ Fig. 1 ] There figure 1 is a simplified view of a device according to the invention for inspecting still-hot glass containers coming out of an example of a manufacturing installation. Fig. 2 ] There figure 2 represents an example of the implementation of an image acquisition system for containers exiting a manufacturing facility and implemented in the inspection device according to the invention. Fig. 3 ] There figure 3 represents another example of the implementation of an image acquisition system for containers leaving a manufacturing facility and implemented in the inspection device according to the invention. Fig. 4 ] There figure 4 is a simplified functional block diagram of an example of a first embodiment of the inspection device according to the invention, implementing a so-called conventional processing of the information contained in the transmitted images and in the infrared radiation images. Fig. 5 ] There figure 5 is a simplified functional block diagram of an example of a first embodiment of the inspection device according to the invention, implementing a segmentation operation on a composite image obtained by matching a transmission image and an infrared radiation image. Fig. 6 ] There figure 6 is a simplified functional block diagram of an example of a first embodiment of the inspection device according to the invention, implementing a convolutional neural network having as input data, matched candidate regions resulting from segmentation operations of a transmission image and an infrared radiation image. Fig. 7 ] There figure 7 is a simplified functional block diagram of an example of a first embodiment of the inspection device according to the invention, implementing two convolutional neural networks having as input data a candidate region resulting from a segmentation operation of a transmission image or an infrared radiation image. Fig. 8 ] There figure 8 is a simplified functional block diagram of an example of a second embodiment of the inspection device according to the invention, implementing a convolutional neural network having as input data a composite image of a transmission image and an infrared radiation image. Fig. 9 ] There figure 9 is a simplified functional block diagram of an example of a second embodiment of the inspection device according to the invention, implementing convolutional neural networks having as input data all or part of a transmitted image and an infrared radiation image, without prior segmentation. Fig. 10A] [Fig. 10B] [Fig. 10C ] There figure 10A is a simulated transmission image showing a defect while the figure10B and the figure 10C These are simulated infrared images representing the same defect if it is respectively a filled cavity or if it is bonded glass. Fig. 11A] [Fig. 11B] [Fig. 11C ] There figure 11A is a transmission image showing a defect while the figure 11B and the figure 11C These are simulated infrared images representing the same defect, whether it is a piece of glass or a grease stain, respectively. Fig. 12A] [Fig. 12B ] There figure 12A is an infrared image while the figure 12B is a transmission image showing an example of possible confusion between stone and broth defects. Fig. 13A] [Fig. 13B ] There figure 13A is an infrared image while the figure 13B is a transmission image showing another example of possible confusion between small defects, bubbling, and stone. Fig. 14A] [Fig. 14B ] There figure 14A is an infrared image while the figure 14B is a transmission image showing another example of possible confusion between broth and stone defects. Description of the implementation methods

[0037] There figure 1 Figure 1 illustrates a device according to the invention for inspecting still-hot glass containers 2 exiting a manufacturing or forming installation 3 of any type known per se. The inspection device 1 is designed to detect, for each container, whether the container has a defect and to identify, for a container with a defect, a type of defect from a family of defects.

[0038] Upon exiting the manufacturing installation 3, the containers 2, such as glass bottles or flasks in the illustrated example, are at a high temperature, typically between 300°C and 600°C. As is known, the containers 2, freshly formed by the installation 3, are taken up by an outfeed conveyor 5 to form a line of containers, in the illustrated example, placed successively on the outfeed conveyor. The containers 2 are transported in a line by the conveyor 5 in a specific direction to be conveyed successively to different processing stations, and in particular to an annealing arch. Upstream of this arch is a surface treatment hood 6, which generally constitutes the first processing station after forming.Advantageously, the inspection device 1 according to the invention inspects the still hot containers upstream of the first surface treatment station, namely the surface treatment hood 6.

[0039] The manufacturing facility 3 is known in itself and an example will be described succinctly only to allow an understanding of the interaction between the inspection device 1 according to the invention and the manufacturing facility 3.

[0040] The manufacturing installation 3 includes a production computer 7 for monitoring the various functionalities of the forming installation 3. Typically, the manufacturing installation 3 comprises several distinct forming sections operating in parallel and successively producing at least one glass container. In the example of the IS machine, the different distinct forming sections each include at least one blank mold receiving a glass parison and at least one finishing mold. It is possible to identify, in a known manner, the forming section, the blank mold, and the finishing mold from which each container 2 originates, as the order in which the containers are produced is known for a given production run.

[0041] The inspection system 1 according to the invention comprises a transmission image acquisition system 10 of the containers 2 and infrared radiation images 10 of the containers 2, and an electronic data processing unit 11 connected to the acquisition system 10. This electronic data processing unit 11 is a computer system of any type comprising computers, external peripherals (display unit, keyboards, etc.), programs, databases, etc. This data processing unit 11 is connected to the production computer 7 in order to receive, if necessary, time-series information from the production computer enabling the association of the containers 2, their images, and their detected defects with the mold number or the forming cavity. Typically, the operation of the image acquisition system 10 is synchronized with the operation of the container forming cavities.Furthermore, this information processing unit 11 transmits identified defects and measurements to the production computer 7, so that the production computer can automatically derive adjustment information for at least one control parameter of the manufacturing installation 3. Such adjustment of the control parameters is carried out manually or automatically. Finally, the information processing unit 11 is connected to an ejector to control the ejection of containers identified as defective, and / or to a display unit to present an operator with the identified defects, transmitted images, and infrared radiation images of the containers.

[0042] The acquisition system 10 allows observation of each container 2, according to two modalities: infrared emission from a hot container and the transmission of light passing through the same container. The acquisition system 10 for transmission images of the containers and infrared radiation images of the containers can be implemented in any suitable manner. According to the example illustrated in the figure 1 , the acquisition system 10 comprises on the one hand, a camera 13 sensitive to infrared radiation emitted by the containers 2 and equipped with a lens 13a and on the other hand, a source 14 of light passing through the containers and a camera 15 sensitive to light transmitted by the containers and equipped with a lens 15a.

[0043] According to an advantageous embodiment, the acquisition system 10 includes a light selection system for the light emitted by the light source 14, positioned to eliminate the light emitted by the source 14 from the radiation captured by the infrared-sensitive camera 13. In other words, the acquisition device 10 is configured so that the infrared-sensitive camera 13 captures only the infrared radiation from the inspected container. This objective can, of course, be achieved in various ways.

[0044] For example, the infrared radiation image of a container is acquired when the light source 14 is switched off. According to another implementation example, the infrared radiation image is acquired along an observation direction such that, with the infrared radiation from the container, the light emitted by the light source 14 is not captured, as in the example illustrated in the figure 1 According to another embodiment, the light source 14 does not emit light within the sensitivity spectrum of the infrared-sensitive camera sensor 13. According to a feature of the invention, the emission spectrum of the light source 14 is in a wavelength range below 0.8 µm, and the infrared-sensitive camera sensor 13 captures infrared radiation in a wavelength range above 0.8 µm.

[0045] The acquisition system 10 may also include optical filters so that the infrared-sensitive camera 13 captures only the infrared radiation from the inspected container. These optical filters can be mounted anywhere between the light source 14 and the infrared-sensitive camera 13.

[0046] There figure 2 This illustrates another embodiment in which the acquisition system 10 comprises a light source 14 illuminating the containers 2, a sensor 13b sensitive to the infrared radiation emitted by the containers 2, and a sensor 15b sensitive to the light emitted by the light source 14 and transmitted by the containers. The acquisition system 10 also includes a common optical lens 18 for recovering the infrared radiation emitted by the containers and the light transmitted by the containers. This optical lens 18 is associated with an optical separation and filtration system 19 to eliminate the light emitted by the light source 14 from the radiation received by the infrared-sensitive sensor 13b.

[0047] There figure 3 This illustrates another embodiment of the acquisition system 10, which includes a camera with a common lens 18 for capturing both the infrared radiation emitted by the containers and the light transmitted through them. The acquisition system 10 also includes two juxtaposed linear sensors located in the focal plane of the camera. One sensor 13b is sensitive to infrared radiation, while the other sensor 15b is sensitive to the light emitted by the source 14. The linear sensors 13b and 15b are arranged aligned with the axis of symmetry of the containers, i.e., vertically. The acquisition system 10 also includes a system for selecting the light emitted by the light source 14, positioned to eliminate the light emitted by the source 14 from the radiation captured by the infrared-sensitive sensor 13b.

[0048] The acquisition system 10 takes, for each container 2, one or more transmission (It) images and infrared (Ir) images, each of these images being two-dimensional. The acquired images are processed by the information processing unit 11 configured to include: * a system for analyzing at least one transmission image (It) and at least one infrared radiation image (Ir), * a system for matching at least one region of a transmission image and at least one region of at least one infrared radiation image, * a classifier of the container, from at least one region of at least one transmission image and at least one region of at least one infrared radiation image, matched, in order to identify for a container, at least one type of defect.

[0049] Information processing unit 11 is thus adapted to implement an inspection process for detecting defects on containers while ensuring the identification, for a given container, of a type of defect within a family of defects. As can be seen from the figures 4 à 9 The inspection method according to the invention consists, for each container, of implementing an acquisition operation Act of at least one transmission image It of the container 2 illuminated by the source 14 of light passing through the container, and an acquisition operation Acr of at least one infrared radiation image of the container. This image acquisition operation may, of course, cover all the containers or only some of them.

[0050] The method according to the invention then consists of performing analysis operations on at least one transmission image and at least one infrared radiation image. The method according to the invention further consists of performing a matching operation or step MC of at least a portion of the transmission image It and at least a portion of the infrared radiation image Ir. The method then consists of performing a classification step CI, based on the information contained in at least one matched transmission image and at least one infrared radiation image, in order to identify, for a container, at least one type of defect Dk.

[0051] According to a first embodiment implemented in the examples of implementation of the figures 4 à 7 The analysis operations aim to process images in such a way as to extract, if a visible feature or object is present, a region corresponding to that object. An object in a digital image is generally a set of connected pixels sharing a common property not possessed by neighboring sets. An object is therefore enclosed by a closed boundary and is recognized as such solely based on image properties: grayscale level, etc.

[0052] An object corresponds to an area or region of an image that potentially contains a defect. A region containing an object, also called a candidate region, presents an object that can be classified as belonging to one of a possible object classes, including defect types. The list of object classes is, for example, as follows: mark of the joint of the container forming mold, shadow, decoration, coat of arms, code, which are not defects; fold, river, rub, orange peel or crevice which are surface defects; fin, trapezoid, blocked neck, which are form defects; crack, inclusion, boil, stone, bubble, which are internal defects in the material; thin which is an area of ​​poor glass distribution.

[0053] According to this more precise approach, the process according to the invention aims to classify each container into at least one class belonging to a family of classes, some of which include types of defects.

[0054] According to the invention, the classification of containers is achieved through the classification of their images or image portions in transmission light (It) and infrared radiation (Ir). It should be noted that a single container frequently exhibits several different defects. Clearly, according to the invention, a first region of the container can be classified based on a first portion of its transmission light (It) image and a first portion of its infrared radiation (Ir) image corresponding to this first portion of the container. Similarly, for the same container, a second region of the container can be classified based on a second portion of its transmission light (It) image and a second portion of its infrared radiation (Ir) image corresponding to this second portion of the container.

[0055] According to the invention, the classification of containers, images, or portions of images aims to ensure production safety and quickly correct process errors. This advantage is particularly significant in the event of a critical defect known as a trapezoid or seesaw defect. This defect is a glass thread inside the container, connected at its ends to the inner wall. The most typical trapezoids run completely through the container, with a loose, downward-curving arc shape. This defect can break, leading to glass fragments in the bottled liquid. It poses a danger to the consumer. Consequently, it is considered critical, and containers with a trapezoid must never be delivered. This classification is successfully achieved through cold or hot inspections, using visible or infrared technology.However, to allow for adjustments to the manufacturing process, it is necessary to detect and identify the defect, preferably while the product is still hot. While this defect is detected because it exhibits a particular characteristic in the images, it is often highly variable: sometimes the wire is missing, and only one or two of the fasteners are visible; sometimes it lacks the typical arc shape, and so on. According to the invention, by classifying the two images based on transmission and infrared radiation modalities, trapezoidal defects can be identified because at least one transmission image (generally two are taken from different viewing angles) accurately reveals the needle-like shape of the defect at the fasteners, while the infrared radiation image indicates an excess of glass. Indeed, infrared radiation, which penetrates the glass and is sensitive to its thickness, is preferred.An operator can therefore be immediately informed and act upon, and / or act automatically upon, the manufacturing process, for example by correcting the roughing temperature or the inverter movement. According to the invention, the trapezoid is therefore part of the list of possible object classes.

[0056] Another illustration of the value of identifying defects through classification is that small inclusions, particularly ceramic ones, are often confused with small air bubbles in one observation method. This confusion is problematic because these defects do not have the same severity and do not have the same cause in the manufacturing process. As will be seen later in the figures 12A-12B And 13A-13B The classification based on images according to the two modalities in transmission and infrared radiation allows them to be distinguished and therefore to act correctly on the process, for example in the case of air bubbles or vapor, to improve the refining of the glass or the lubrication of the scissors and the conduits called "deliveries".

[0057] The analysis of transmission (It) and digital infrared (Ir) images employs operations known in themselves, notably filtering and segmentation, performed in such a way as to extract all regions containing an object. Thus, the method implements an SRt operation for segmentation and detection of candidate regions on transmission (It) images and an SRr operation for segmentation and detection of candidate regions on infrared (Ir) images. figures 4 , 6 And 7 ). According to the example of implementation illustrated in the figure 5 , an SR operation of segmentation and detection of candidate regions is performed on a composite image of a transmission It image and an infrared radiation Ir image as will be explained later in the description.

[0058] A segmentation operation typically involves dividing the image into regions or segments, that is, assigning each pixel a region. This segmentation aims to identify candidate regions within each image through filtering, thresholding, edge tracking, and other techniques, generally, but not necessarily, to measure parameters that characterize these regions. This image segmentation is performed using a filtering method adapted to the specific modality, i.e., transmission (It) images and infrared (Ir) images.

[0059] These SRt, SRr, and SR segmentation operations allow the detection of candidate image regions defined by their contour limited to the object. These candidate image regions, RTC, RRC, and RCC, are respectively in transmission, infrared radiation, or a combination of transmission and infrared radiation. It is also possible that these SRt, SRr, and SR segmentation operations can detect candidate image regions defined by their rectangle enclosing the object, RTE, RRE, and RCE. These candidate image regions are respectively in transmission, infrared radiation, or a combination of transmission and infrared radiation.It is also possible that these SRt, SRr,SR segmentation operations allow the detection of candidate image regions defined by its expanded rectangle framing the object RTL, RRL, RCL, so as to take into account the context of the object, these candidate image regions being respectively in transmission, in infrared radiation or composite in transmission and infrared radiation.

[0060] The method according to the invention implements a MC operation for matching candidate regions of transmission images or candidate regions of infrared radiation images with corresponding regions of infrared radiation images and transmission images, respectively, based on their position on the container. This matching can also apply to candidate regions of transmission images with candidate regions of infrared radiation images.

[0061] This MC matching operation aims to ensure that regions in the two images correspond by comparing their respective positions on the container. In the most general case, a geometric transformation is determined from one image to the other, which allows, starting from a region or pixel in one container image, the location of a region or pixel in the other image corresponding to the same region or elementary part of the container. The geometric transformation can be of any type necessary and may include, for example, translation / rotation, anamorphosis, scaling, etc.

[0062] In one embodiment, a pixel-by-pixel mapping is performed between two images or image regions of a container. To do this, the geometric transformation is determined for all pixels. It is also possible to calculate, for one of the two images or image portions, a transformed image that is superimposable on the other image or image portion. The geometric transformation, along with interpolations (for example, bilinear interpolations) of the pixel values, is then applied to all pixels in the region concerned.

[0063] In cases where the transmitted and infrared images correspond pixel by pixel due to the image acquisition system, the matching is direct. In this case, the acquisition device must be constructed with great precision so that the camera sensors have the same field of view, magnification, viewing direction, and resolution in pixels per mm. Thus, the matching is already achieved because the pixels of each transmitted and infrared image correspond to the same elementary portion of the container's surface. Of course, any deviation from this ideal situation can be compensated for by matching using a suitable geometric transformation.

[0064] According to another embodiment, regions whose midpoints or centers of gravity are close on the container are matched; that is, they correspond or are adjacent under the geometric transformation. Alternatively, regions whose rectangles enclosing the RTE, RRE, RCE object, or enlarged rectangles enclosing the RTL, RRL, RCL object, intersect or overlap on the container within a certain proportion of their surface area are matched.

[0065] Also, this MC matching can be done either from candidate regions to candidate regions or from pixel to pixel as in the illustrated implementation example.

[0066] According to the examples of implementation illustrated in figures 5 And 8A composite image (CI) is created by matching a transmission (IT) image or portion thereof with an infrared (IR) image or portion thereof. The process thus ensures, through matching, a fusion of the transmission and infrared images to obtain the composite image (CI). The matching operation (MC) can involve all transmission and infrared images or only parts of them. This MC matching is performed pixel by pixel, as explained previously. Each pixel pc(x,y) with coordinates x, y in the composite image is assigned a value that depends on the transmission obtained from the IT image and the infrared radiation obtained from the IR image.This value is, for example, either a 16-bit scalar pc(x,y) with 8 transmission bits and 8 infrared radiation bits, or a vector vc(x,y) whose components {vt(x,y), vr(x,y)} are each a transmission scalar and an infrared radiation scalar. The simplest method is to directly record the value of a pixel in the transmission (It) image and a corresponding pixel in the infrared (Ir) image. However, it is possible to construct each composite pixel from a combination of several neighboring pixels in the images, or from interpolated values.

[0067] According to the examples of implementation illustrated in figures 4 , 6 And 7The matching process is performed on a candidate region-to-candidate basis. It is possible to register one image to the other to align candidate regions in two images of different modalities. It is also possible to directly search for candidate regions located within the same area of ​​the container.

[0068] The method according to the invention aims to determine classification criteria in radiation and transmission, which include transmission criteria that take into account the characteristics ti (in number n) of transmission images and infrared radiation criteria that take into account the characteristics ri (in number m) of infrared radiation images, and / or composite criteria that take into account characteristics ci (in number q) combining transmission and infrared radiation images in a logical or mathematical manner. These characteristics are, for example, characteristics of position, size, shape (concavity, perimeter, surface area, etc.) or photometry (average level, contrast, variance, textures, etc.).

[0069] According to the examples of implementation illustrated in figures 4 And 5The method according to the invention implements an ECt, ECr, EC operation for extracting classification criteria. According to the embodiment illustrated in the figure 4 The method implements an ECt operation for extracting classification criteria for the candidate region RTC, RTE, RTL from transmission images, enabling the definition of an n-dimensional vector Ct representing n ti characteristics obtained from the It radiation image for each candidate region. Similarly, the method according to the invention implements an ECr operation for extracting classification criteria for the candidate region RRC, RRE, RRL from infrared radiation images, enabling the definition of an m-dimensional vector Cr representing m ri characteristics obtained from the Ir radiation image for each candidate region.

[0070] It should be noted that according to the example of implementation illustrated in the figure 4 , the MC matching operation of candidate regions RTC, RTE, RTL of transmission images with candidate regions RRC, RRE, RRL of infrared radiation images allows obtaining a vector Cc of dimension n+m+q representing n+m+q characteristics ti, ri and ci obtained for each candidate region matched between transmission images and infrared radiation images.

[0071] In the example of implementation illustrated in the figure 5 The method according to the invention implements an EC operation for extracting transmission and radiation classification criteria for the composite candidate region in transmission and infrared radiation RCC, RCE, or RCL, obtained after the SR segmentation operation. This extraction operation makes it possible to obtain a vector Cc of dimension n+m+q representing n+m+q characteristics ti, ri, and ci obtained for each candidate region matched between transmission and infrared radiation images.

[0072] In the examples of implementation of the figures 4 And 5 The classification criteria are determined by a preliminary analysis, namely industry knowledge or statistical studies. Image analysis algorithms determine the position, size, shape, and photometric characteristics (ti, ri, ci). Examples of implementation are illustrated in the following sections. figures 6 And 7The classification criteria are determined by supervised learning by being embedded in trained neural networks CNN, CNN1, CNN2. The training set contains pairs of candidate regions (RTC, RRC), (RTE, RRE), (RTL, RRL) according to the two inspection modalities, with a label representing an object class from a list of possible classes such as defect types.

[0073] Using predetermined transmission and radiation classification criteria, the process classifies defects and, consequently, the containers bearing these defects. The classification operation determines the object class Dk of the candidate region or container from among p possible classes D1, D2, ...Dp. If a candidate region is found in only one of the two images according to a first method, an analysis of the defect is performed according to the criteria associated with the image type used, but also taking into account criteria associated with the other method: an analysis is performed based on the fusion of criteria associated with both image types. The principle of the invention relies on considering both inspection methods to provide additional and reliable information for classifying objects or containers, and therefore for identifying defects.

[0074] The classification decision assigns a class Dk to the container from among p possible classes. The p classes are primarily types of defects and include, in particular, bubbles, folds, rivers, rubs, orange peel, crevices, fins, trapezoids, blocked necks, bubbles, stones, mold seam marks, shadows, decorations, coats of arms, codes, thin lines, etc. Several classes may be defined for the same defect if that defect presents varied shapes, such as trapezoid 1 and trapezoid 2. According to a preferred embodiment, the object of the invention is to classify each container according to at least one object class from a list of possible classes containing at least several types of defects, the list of possible classes including at least: no defect, trapezoid, inclusion, and bubble.

[0075] In the examples of implementation illustrated in figures 4 And 5The CI classifier can be, for example, a Support Vector Machine (SVM), a Bayesian classifier, or a Neural Network (NN). The CI classifier performs classification using predetermined transmission and radiation classification criteria. The CI classifier is trained using supervised learning methods that determine classifier parameters from a set of objects or images of known class, called the training set, and usually also a test set. The training set comprises pairs of regions, preferably pairs of characteristic vectors Ct, Cr associated with a defect type, i.e., {Ct, Cr, Dk}. Typically, the supervised learning classifier is trained on a training database consisting of a set of records, each containing, for each observed example container: at least one radiation image of the container, at least one transmission image of the container and at least one label assigning to the example container at least one object class from a list of possible classes such as defect types, or at least one radiation image region of the example container, at least one transmission image region of the example container and at least one label assigning to the corresponding region of the example container at least one membership class from a list of object classes such as defect types.

[0076] According to the example of implementation illustrated in the figure 6 , the classifier is a convolutional neural network (CNN) having as input data a pair of candidate regions (RTC, RRC), (RTE, RRE) or (RTL, RRL), obtained after the MC matching operation.

[0077] According to the example of implementation illustrated in the figure 7 The classifier comprises a first convolutional neural network, CNN1 (Convolutional Neural Network), which takes as input a candidate region in transmission (RTC, RTE, RTL) obtained after the SRt operation of segmentation and candidate region detection on transmission (It) images. The classifier also comprises a second convolutional neural network, CNN2 (Convolutional Neural Network), which takes as input a candidate region in infrared radiation (RRC, RRE, RRL) obtained after the SRr operation of segmentation and candidate region detection on infrared (Ir) images.The first convolutional neural network CNN1 and the second convolutional neural network CNN2 each work in parallel respectively on a candidate region in transmission (RTC, RTE, RTL) and on a candidate region in infrared radiation (RRC, RRE, RRL), these two regions according to the two modalities, being associated by the MC matching operation according to the techniques explained above.

[0078] The outputs of the first convolutional neural network (CNN1) and the second convolutional neural network (CNN2) are the input data for a classifier, for example, a System Vector Machine (SVM), Random Forest, Bayesian, or preferably a Neural Network (NN), allowing classification according to both modalities. The outputs of the first convolutional neural network (CNN1) and the second convolutional neural network (CNN2) are, for example, class membership hypotheses, but they can also be more complex data with vectors of dimensions greater than p classes. The neural network training set contains pairs of candidate regions (RTC, RRC), (RTE, RRE), (RTL, RRL) according to the two inspection modalities, labeled with an object class from a list of possible classes, such as defect types.

[0079] According to a second embodiment implemented in the examples of implementation of the figures 8 And 9The image analysis operations do not aim to extract a candidate region but rather to consider all or part of the transmission (IT) and infrared (Ir) images, without prior segmentation. According to these two implementation examples, the analysis operations rely on the use of neural networks. If only parts of the images are analyzed, these parts preferably correspond to one or more regions of interest on the container, such as the ring, neck, shoulder, body, mouth, or a right or left half of the side, or an area containing engravings.

[0080] In the example of implementation illustrated in the figure 8 A composite image (CI) operation is performed to match a transmission image (It) with a radiation image (Ir) to obtain a composite image (CI). The composite image (CI) is obtained by merging at least one radiation image with at least one transmission image of a container, or by merging regions of at least one radiation image with corresponding regions of at least one transmission image.

[0081] The transmission (It) image and the radiation (Ir) image are captured during the Act and Acr acquisition operations performed by the image acquisition system 10, as explained in the description above. This pixel-to-pixel MC matching of the images is performed as explained in the implementation example of the figure 5 This composite IC image is used as input data for a convolutional neural network (CNN) that is trained to consider position, size, shape, and photometry features that are significant for the intended classification. The training set contains composite IC images or regions of composite images, each labeled with an object class from a list of possible classes, such as defect types. Transmission and radiation classification criteria are incorporated into the weights resulting from the training and defining the CNN. It is worth noting that, unlike the examples of figures 5 And 6The segmentation operation is not necessary in this variant because the stages of the convolutional neural network are capable, through learning, of classifying images according to their content without prior segmentation, and of locally determining the position, size, shape, and photometry characteristics that are significant for classification. However, a segmentation operation is possible, for example, by replacing figure 5 feature extraction EC and CL classifier by a convolutional neural network (CNN) based classifier.

[0082] In the example of implementation illustrated in the figure 9 The transmission image (It, in part or in whole) captured by the image acquisition system 10 is used as input data for a first convolutional neural network, CNN1, which is trained to determine the position, size, shape, and photometry features that are significant for the intended classification. The infrared radiation image (Ir, in part or in whole) captured by the image acquisition system 10 is used as input data for a second convolutional neural network, CNN2, which is trained to determine the position, size, shape, and photometry features that are significant for the intended classification.

[0083] The first convolutional neural network (CNN1) and the second convolutional neural network (CNN2) each work in parallel on two candidate images in each modality. The outputs of the first CNN1 and the second CNN2 serve as input data for a classifier, for example, a System Vector Machine (SVM), Random Forest, Bayesian, or preferably a Random Neural Network (NN), which classifies the containers according to the two modalities. It should be noted that the two candidate images in each modality on which the first CNN1 and second CNN2 work are associated with a matching operation.

[0084] The outputs of the first convolutional neural network CNN1 and the second convolutional neural network CNN2 are, for example, class membership hypotheses, but they can be more complex data with vectors of dimensions greater than the number p of classes. The neural networks' training set contains pairs of images under the two inspection modalities, labeled with an object class from a list of possible classes, such as defect types. The transmission and radiation classification criteria are taken into account in the weights resulting from the training and defining the neural networks CNN1, CNN2, and NN. It should be noted that, unlike the example of the figure 7 The segmentation operation is not necessary in this variant because the stages of the convolutional neural network are able, through learning, to classify images according to their content without prior segmentation, and to locally determine the position, size, shape and photometry characteristics that are significant for classification.

[0085] The object of the invention is advantageously exploited within manufacturing facilities to enable better detection and categorization of defects present within still-hot formed containers. Certain defects can be seen, detected, and categorized more easily according to one of the two methods or through a combination of both methods.

[0086] There figure 10A is a simulated image of a defect seen in transmission, at the level of the puncture in a container, while the figures 10B et 10C These are simulated images of the same defect seen in infrared radiation, showing whether it is a filled cavity or glued glass, respectively. This defect appears as a trapezoid enclosed within a less absorbent cavity. It's as if an air bubble has formed between the puncture and its apex, leaving only a strand of glass. It could also be another piece of glass, glued to the container wall or attached to the puncture inside the container. An infrared image could potentially confirm or rule out the presence of a strand of glass, which emits more radiation, and an air bubble, which emits less.

[0087] In the case of a strong contrast change for the transmission modality and a strong contrast change for the infrared radiation modality, it can be concluded that there is an accumulation of matter at that location corresponding to a trapezoid, contained within an air bubble ( figure 10B ). In the case of a strong contrast change for the transmission modality and a weak contrast change for the infrared radiation modality, this involves two pieces of glass glued together at the point of the puncture ( figure 10C ).

[0088] There figure 11A is an image of a defect seen in transmission while the figures 11B et 11C These are simulated images of the same defect as seen in infrared radiation, representing a piece of glass or a grease stain, respectively. It could be a critical defect, such as a trapezoidal shape (a piece of glass inside the container). However, it could also be a grease stain, a superficial defect visible on the container wall. If it is a thick piece of glass inside the container, its infrared radiation will combine with that of the container walls, increasing in intensity. Conversely, if it is a simple grease stain, there should not be a significant difference between the infrared radiation from the container wall and that of the superficial defect.

[0089] In the case of a strong contrast change for the transmission modality and a strong contrast change for the infrared radiation modality, this is an accumulation of matter at that location corresponding to a trapezoid ( figure 11B ). In the case of a strong contrast change for the transmission modality and a weak contrast change for the infrared radiation modality, this is a visible light absorbing defect, probably a grease stain on the surface of the container ( figure 11C ).

[0090] THE figures 12A , 13A , And 14A are real images obtained with an InGaAs infrared camera, while the figures 12B , 13B And 14Bare real images of the same defects seen in transmission, the acquisitions of the container images having been carried out at the exit of the manufacturing machine in a manufacturing plant where cameras are installed according to the inspection device conforming to the invention.

[0091] There figure 12A is an image of a defect seen in infrared radiation while the figure 12B is an image of the same defect seen in transmission. According to the prior art, the defect observed on the transmitted image of the figure 12B This leads to uncertainty as to its classification as either a stone or a broth. According to the invention, the infrared radiation image shows a defect that emits infrared radiation corresponding to the presence of glass, thus resolving the confusion and concluding that a stone is present.

[0092] There figure 13A is an image of a defect seen in infrared radiation while the figure 13B is an image of the same defect seen in transmission. The defect in the transmission image suggests a stone or boulder defect. However, the infrared image shows that the defect emits little light and is less contrasted than in the figure 12A The problem is therefore a broth.

[0093] There figure 14A is an image of a defect seen in infrared radiation while the figure 14B This is an image of the same defect seen in transmission. The defect in the transmission image suggests a boulder defect due to the light transition at the center of the dark spot. However, the infrared image shows that the entire defect radiates, indicating that the defect is a stone.

[0094] These various examples demonstrate the advantage of the invention's use of two inspection methods to improve both the detection and classification of defects. The second inspection method confirms the defect classification made using the first method or refutes it by allowing the defect to be classified in a different category. For defects that are not easily visible in one method, an image obtained using the other method provides additional information for correctly classifying the defects.

[0095] It can happen that the signal from a defect is weak in both modalities considered. It should be noted that the transformation into a composite image can reveal objects that were too weakly contrasted in both modalities, but which, once the fusion is performed, become easier to identify.

[0096] The invention applies to any method for manufacturing glass containers, including bottles, jars, flasks, syringes, ampoules, drinking glasses, jars, and plates. Indeed, in all these manufacturing processes, after forming, there is a lengthy cooling stage for the glass objects, and early inspection and detection of defects is essential.

Claims

1. Method for inspecting glass containers (2) still hot and exiting a manufacturing facility (3) to identify, for a container, a type of defect, the method comprising, for each container: ∘ acquiring at least one transmission image (It) of the container illuminated by a light source (14) emitting light passing through the container, and at least one infrared radiation image (Ir) of the container, ∘ analyzing at least one transmission image and at least one infrared radiation image, ∘ performing a matching of at least a portion of the transmission image and at least a portion of the infrared radiation image, ∘ classifying the container, based on at least one matched transmission image and at least one matched infrared radiation image, to identify at least one type of defect for a container.

2. Inspection method according to claim 1, wherein the container is illuminated by a light source (14) whose emission spectrum is in a wavelength range below 0.8 µm, and the infrared radiation image of the container is acquired in a wavelength range above 0.8 µm.

3. Inspection method according to any of the preceding claims, wherein the infrared radiation image of a container is acquired when the light source (14) is turned off, or wherein the infrared radiation image is acquired in an observation direction such the light emitted by the light source (14) is not captured with the infrared radiation of the container.

4. Inspection method according to any of the preceding claims, wherein, to perform the matching of transmission images and infrared radiation images, the method detects candidate regions in the transmission images and in the infrared radiation images, the method performing, for each container: ∘ a matching of candidate regions (RTC, RTE, RTL) of the transmission images or candidate regions of the infrared radiation images (RRC, RRE, RRL) with the corresponding regions of the infrared radiation images and transmission images, respectively, based on their position on the container, ∘ or a matching of candidate regions of the transmission images with candidate regions of the infrared radiation images.

5. Inspection method according to the preceding claim, wherein the matching comprises a fusion of the transmission images and the infrared radiation images to obtain a composite image (IC), the method performing: ∘ an extraction of classification features from the composite image, expressing classification criteria in transmission and radiation, and ∘ a classification of the container using the classification criteria in transmission and radiation; or the method performing: ∘ a segmentation of the composite images to detect composite candidate regions, ∘ an extraction of classification features from the composite candidate regions, expressing classification criteria in transmission and radiation, ∘ a classification of the container using the classification criteria in transmission and radiation from the composite candidate regions.

6. Inspection method according to any of the preceding claims, wherein: ∘ classification criteria in transmission are extracted from the transmission images, ∘ classification criteria in radiation are extracted from the infrared radiation images, ∘ the container is classified using the classification criteria in transmission and radiation.

7. Inspection method according to the preceding claim, wherein classification criteria in radiation are selected for the infrared radiation images and classification criteria in transmission for the transmission images, and / or composite criteria that logically or mathematically combine features from transmission and infrared radiation images, these classification criteria in radiation and transmission being criteria of position, size, shape, or photometry.

8. Inspection method according to any of the preceding claims, wherein the container is classified by a supervised learning classifier whose input data are: ∘ the classification criteria in radiation and transmission, ∘ or the radiation and transmission images, ∘ or parts of the radiation and transmission images, ∘ or at least one composite image (IC) obtained by fusing at least one radiation image with at least one transmission image of a container or by fusing regions of at least one radiation image with corresponding regions of at least one transmission image.

9. Inspection method according to claim 8, wherein the container is classified by a supervised learning classifier trained on a training database comprising a set of records, each including, for an observed example container: ∘ at least one radiation image of the container, at least one transmission image of the container, and at least one label assigning the example container to at least one object class among a list of possible classes such as defect types, or ∘ at least one radiation image region of the example container, at least one transmission image region of the example container, and at least one label assigning the corresponding region of the example container to at least one object class among a list of possible classes such as defect types.

10. Method according to any of the preceding claims, wherein each container is classified into at least one object class among a list of possible classes including at least: no defect, birdswing, inclusion, bubble.

11. Inspection method according to any of the preceding claims, wherein a step is implemented to take into account at least one detected defect type to deduce adjustment information for at least one control parameter of the manufacturing facility.

12. Device for inspecting glass containers (2) still hot and exiting a manufacturing facility (3) to identify, for a container, a type of defect, the device comprising: • an image acquisition system (10) for acquiring transmission images and infrared radiation images of the containers, • an information processing unit (11) connected to the image acquisition system (10), the information processing unit (11) being configured to include: ∘ an analysis system for at least one transmission image and at least one infrared radiation image of the container, ∘ a matching system for at least one region of a transmission image and at least one region of at least one infrared radiation image of the container, ∘ a classifier for the container, based on at least one region of at least one transmission image and at least one region of at least one infrared radiation image, matched, to identify at least one type of defect.

13. Device according to claim 12, wherein the image acquisition system (10) for transmission and infrared radiation images of the containers: • comprises, on one hand, a camera (13) sensitive to infrared radiation emitted by the containers and equipped with a lens (13a), and on the other hand, a light source (14) emitting light passing through the containers and a camera (15) sensitive to the transmitted light and equipped with a lens (15a), or • comprises a light selection system (19) positioned to eliminate, from the radiation captured by the infrared-sensitive camera, the light emitted by the light source (14), or • comprises: ∘ a light source (14) illuminating the containers, ∘ a sensor (13b) sensitive to infrared radiation emitted by the containers, ∘ a sensor (15b) sensitive to the light emitted by the light source (14) and transmitted through the containers, ∘ a common optical lens (18) for collecting infrared radiation emitted by the containers and light transmitted through the containers, this optical lens (18) being associated with an optical separation and filtering system to eliminate the light emitted by the light source from the radiation received by the infrared-sensitive sensor.

14. Device according to any of claims 12 to 13, wherein the information processing unit (11) is connected: • to an ejector to command the ejection of containers identified as defective, and / or • to a display unit to present to an operator the identified defects, the transmission images, and the infrared radiation images of the containers.

15. Device according to any of claims 12 to 14, wherein the information processing unit (11) is connected to a production computer (7) supervising the manufacturing facility (3) to: • receive temporal information from the production computer allowing the association of containers, their images, and detected defects with the mold number or forming cavity, • transmit to the production computer the identified defects and measurements performed, so that the production computer can automatically deduce adjustment information for at least one control parameter of the manufacturing facility.

Images